Fuel Cell Manufacturing: An AI Approach
Mussawar Ahmad1 (mussawar.ahmad@warwick.ac.uk), Prof Robert Harrison1, Dr James Meredith2, Dr Axel Bindel3, Dr Ben Todd4
1
Automation Systems Group, International Manufacturing Centre, WMG, University of Warwick, Coventry, CV4 7AL, United Kingdom
2
Mechanical Engineering, University of Sheffield, Garden Street, South Yorkshire, S10 2TN, United Kingdom
3
HSSMI, CEME Campus, Marsh Way, Rainham, RM13 8EU, United Kingdom
4
Arcola Energy, 24 Ashwin Street, London, E8 3DL, United Kingdom
2. Research Questions
1. Introduction
Fuel cells are electrochemical devices which convert H 2 and O2 to electricity.
They are used in portable, transport and stationary power applications (Fig. 1).
There are several barriers to mass production, the most commonly quoted
being a lack of hydrogen infrastructure and high product cost as compared
to incumbent technologies (Fig.2). This research focuses on reducing cost by
better managing and using manufacturing assembly knowledge using
knowledge representation (KR) which is a field of artificial intelligence (AI).

What fuel cell assembly knowledge should be captured?

What can and should be done with this knowledge?

How can you add to this knowledge base as the technology develops?
3. Method
Knowledge is the ability to understand information and then make decisions.
In this research we want to understand product information and make a
decision on what manufacturing equipment to use. This is done using KR
which is composed of two components (i) a knowledge base (ii) a set of rules
or axioms from which inferences can be made (Fig. 2)
PEM stands for proton
exchange membrane.
The membrane and
electrodes form the
heart of the fuel cell
50% efficiency
compared to 25% of
internal combustion
engine
Only emission is
water
Fig 2. Fuel Cell cost reduction and knowledge focus
5. Case Study
Fig 1. Fuel Cell applications and operating principle
4. Model
E.g Move, transport, check, rotate, grip, release etc...This
information is expert domain knowledge residing in the heads
of process planners
Membrane electrode assembly (MEA)
is the modelled product domain
The equipment required
to meet the requirements
of the product and
process. This could be
robots, grippers,
conveyors and operators.
This is what the knowledge associated
with that assembly looks like
This table is generated
when querying for appropriate equipment
Geometric information e.g.
dimensions, weight AND nongeometric information e.g.
function, mating conditions.
This type of data is typically
stored in the computer aided
design (CAD) document or
accessible to the product
designer
Reasoner correctly selects a robot and
vacuum gripper as the appropriate assembly
equipment to assembly a MEA based on
component characteristics
Fig 3. Model overview
6. References
6. Conclusions and Further Work
 The concept has been proved—equipment can be generated and the assembly sequence model works
 As an ontology has been used, the model is extensible and scalable allowing for the addition of more
information in the future
 But it is a time consuming process! Therefore the existing data sets need to be exploited to automate
the process
[1] J. L. Nevins and D. E. Whitney, “Concurrent design of product and processes,” McGraw-Hill,
New York, 1989
[2] U. Rembold, C. Blume, and R. Dillmann, “Computer- integrated manufacturing technology
and systems,” Marcel Dekker, New York, 1985
[3] S. S. F. Smith, “Using multiple genetic operators to reduce premature convergence in genetic
assembly planning,” Computers in Industry, Vol. 54, Iss. 1, pp. 35–49, May 2004